Phase diagram of CaSiO3 perovskite from deep potentials with ab initio accuracy*

Thermodynamic properties of minerals are crucial in revealing the structure and properties of planetary interiors. CaSiO3 perovskite (CaPv) is believed to be the third most abundant mineral in the Earth’s lower mantle. A previous study suggested that its tetragonal-to-cubic transition can occur at lower-mantle conditions, potentially causing seismic anomalies, particularly in the Large Low-Shear-Velocity Provinces (LLSVPs) (Thomson et al., Nature 572, 643, 2019). However, ab initio methods for describing the tetragonal-cubic phase transitions in CaPv were limited by their high computational cost and difficulties in addressing a nearly second-order phase transition. We utilized a machine-learning method and novel thermodynamic integration techniques to investigate the phase diagram of CaPv at temperatures ranging from 300-3,000 K and pressures up to 130 GPa. Our simulations using a deep-learning potential with ab initioaccuracy indicate that the tetragonal-to-cubic transition in CaPv happens at ~1,000 K below mantle temperatures. Therefore, the lower mantle should have only cubic-CaPv. This suggests LLSVPs are caused by other mechanisms.